Algorithm

It's built from steps and conditions

Two things belong in every algorithm: an order — what to do in which sequence — and a condition — what happens, and when, that changes the path. In "once it boils, add the noodles," the first half is the condition and the second half is the action.

Leave a gap and it stops being a procedure. "Season it to taste" works fine for a person, but a computer has no way to know how much of what to add. Writing an algorithm means filling in every one of those vague spots, with nothing left over.

It also needs a defined stopping point. A procedure that repeats the same step needs to say when to quit, or it becomes a loop that never ends.

Changing the order changes the result

The same ingredients give different results depending on order — whether the seasoning goes in before or after the water boils changes the taste. In an algorithm, order isn't decoration; it's the core of the procedure.

The same holds for what a computer does. Filtering data and then sorting it can take a very different amount of time than sorting first and then filtering. Even when the final result is identical, one order can finish in seconds while the other takes hours.

The same problem has more than one solution

Take a list of names and ages that needs sorting alphabetically. You could scan through from start to finish, swapping entries as you go, or split the list in half, sort each half, and merge them back together. Both land on the same sorted list, but as the list grows longer, the time gap between the two methods stretches enormously.

That's why choosing an algorithm isn't only about the result — it's about how long it takes and how much room it needs. A short list favors the simple method; a very long one favors the faster, more complicated one. The right answer depends on the situation.

In AI, algorithm means the training method

In machine learning, algorithm means the procedure for how to adjust the numbers inside a model, based on the data. Measure how far off it is, decide which direction and by how much to adjust, adjust, and measure again — the same fixed rule, repeated.

What's left once that procedure finishes is a result with the numbers settled into place. That's the model. A recipe isn't the noodles, and a training algorithm isn't the model, either. The algorithm is what a person writes down in advance; the model is what comes out of applying that procedure to data.

Which is why the same algorithm produces a completely different model depending on what data you feed it, the same way the same recipe produces a different dish depending on what ingredients go in.

"The algorithm" in the news means something broader

When someone says "the algorithm put this video in front of me," they don't mean a single procedure. It's shorthand for several stages of deciding what to show, a handful of trained models, and rules the company running the service set on top of all of it.

The word sounds mysterious because none of that is visible from outside. Nobody outside can tell what standard was used to pick something, and often the people who built it can't fully explain the result either — which is exactly why arguments break out over why a recommendation turned out the way it did.